When tuning a Spark job that processes a dataset with uneven data distribution (skewed data), which configuration setting is most effective for ensuring the workload is evenly distributed across all cluster nodes?